A few days ago, the Chinese company Moonshot AI released Kimi K3, the largest open-weight model to date, featuring a set of 2.8 trillion parameters. The launch of Kimi K3 has already sparked debate in Silicon Valley, as it narrows the gap with top AI models from US companies: Anthropic’s Claude Fable and OpenAI’s GPT-5.6 Sol. Now, Moonshot AI is making Kimi K3 available for public download.

When the launch of Kimi K3 was announced on July 16, Moonshot AI stated that the model’s open weights would be made public on July 27.

Run Kimi K3 locally

To put this in context, the term “open weights” refers to the set of parameters used to train an AI model—in this case, Kimi K3.

In other words, once the open weights are available for download, you will be able to run Kimi K3 locally on your device and modify it to suit your needs, without worrying about token costs. These are the same token costs that have led major companies like Uber and Walmart to rethink how their employees use AI tools such as Claude Code.

Given that Kimi K3 is a massive model with over 2.8 trillion parameters, you will likely need enterprise-grade infrastructure to run it, rather than simply using your laptop. It is worth noting that, at the time of writing, the open weights were not yet available for download AI’s official website. Are OpenAI and Anthropic worried?

The release of the open weights for Kimi K3 comes amidst a growing divide in Silicon Valley as Chinese AI models improve their performance. As models like Kimi K3 and Zai’s GLM 5.2 approach the level of cutting-edge US models, companies like OpenAI and Anthropic may have begun to worry, given that the cost for users to run these Chinese models is much lower.

According to a famous publication report, Anthropic and OpenAI have been lobbying the Trump administration to impose stricter restrictions on Chinese open-weight AI models. These US startups argue that Chinese models could raise concerns regarding intellectual property theft and national security.

It is worth noting that Anthropic has previously accused Chinese AI labs of using a distillation technique with its models to train their own. Distillation is essentially a process in which a smaller AI model learns by analysing the responses of a larger model.

Michael Kratsios, a White House advisor, has also accused Moonshot AI of distilling Anthropic’s models to train Kimi K3.

At the same time, there are voices openly supporting open-weight AI models. On July 24, 25 US tech companies signed a joint letter titled “Open Weights and US Leadership in AI,” urging Washington to avoid premature restrictions on downloadable AI models.

Signatories included Nvidia, Microsoft, Meta, IBM, Dell, and Palantir. Nvidia CEO Jensen Huang—who had never posted on X before—used his first post on the platform to share the letter. OpenAI CEO Sam Altman added his signature after the document was made public, bringing the total number of signatories to 50. Anthropic remains absent from the list of signatories. Chinese open-weight models have gained popularity because they are much more cost-effective to run than closed AI models from the United States. Following the public launch of Kimi K3, Moonshot AI had to suspend new registrations, as its graphics processing units (GPUs) had reached capacity limits.